六月麦茬

Hi, Thanks for your advise. I've found the problem. the shape of parameters from Tensorflow is [1, 1, 1, 1, n], mismatch the size from PyTorch ([n]). I've fixed the...

Yes, you reminded me ! I updated my Pytorch from 0.3 to 0.4.1 recently, and yesterday I found the similar problem in another program. Thank you for your help even...

Why not? Are you sure the problem comes from the open source? I can run it with GPU, and other people also can do it.

在我实验中没有遇到这种情况，你可以先熟悉下代码。相信你自己可以解决这个问题的。我现在一直忙于其他项目，目前脱不开身。抱歉……

Sorry, I don't concat previous two messages as the input of generator, because I don't have enough three or more turns dialogue. If you have the related corpus, I think...

我是看D的loss值（accuracy）和G 的loss值。还有在最后的模型手动测试对比。您说的perplexity，在之前的实验中也有用到，但当时感觉和G的loss值差不多，perplexity降低了也不能体现聊天模型就一定会更好。所以我就没用。不过可以加上试试。谢谢您的提醒。gan训练模型时，G的预训练要把握好，一般预训练到能做 一般回复就好。al_train的训练过程中，Dloss值一般是趋于下降，而G的loss值先稍微上扬，在逐渐下降，不知道你训练中是否这样……

The author of this paper have opened the code in lua language, and he also provided the training data. You can get the corpus in there.

Sorry，I haven't finished this project. Somewhere in the code may have bugs.

Sorry, I am studying on the zero-shot learning recently. I am interning in Hitachi now, so I have no time to do that, I will do my best to complete...

Sorry，I don't describe the type in details。in type 2，encoder_input_a is not a item,it‘s a encoder data set，so, firstly we get the encode data set encoder_input_a, and get a encode sentence...